异步通信指令详解（TPUT_ASYNC / TGET_ASYNC / BuildAsyncSession）
【免费下载链接】pto-isaParallel Tile Operation (PTO) is a virtual instruction set architecture designed by Ascend CANN, focusing on tile-level operations. This repository offers high-performance, cross-platform tile operations across Ascend platforms.项目地址: https://gitcode.com/cann/pto-isa
TPUT_ASYNC — 异步远程写
启动 GM→GM DMA 传输，立即返回AsyncEvent。
template <DmaEngine engine = DmaEngine::SDMA, typename GlobalDstData, typename GlobalSrcData, typename... WaitEvents> AsyncEvent TPUT_ASYNC(GlobalDstData &dst, GlobalSrcData &src, const AsyncSession &session, WaitEvents&... events);TGET_ASYNC — 异步远程读
启动远端 GM→本地 GM DMA 传输。
template <DmaEngine engine = DmaEngine::SDMA, typename GlobalDstData, typename GlobalSrcData, typename... WaitEvents> AsyncEvent TGET_ASYNC(GlobalDstData &dst, GlobalSrcData &src, const AsyncSession &session, WaitEvents&... events);BuildAsyncSession — 构建异步会话
SDMA 构建（默认）
template <DmaEngine engine = DmaEngine::SDMA, typename ScratchTile> bool BuildAsyncSession(ScratchTile &scratchTile, __gm__ uint8_t *workspace, AsyncSession &session, uint32_t syncId = 0, const sdma::SdmaBaseConfig &baseConfig = {sdma::kDefaultSdmaBlockBytes, 0, 1}, uint32_t channelGroupIdx = sdma::kAutoChannelGroupIdx);| 参数 | 说明 |
|---|---|
scratchTile | 用于 SDMA 控制元数据的 UB scratch tile（非数据负载），推荐Tile<TileType::Vec, uint8_t, 1, comm::sdma::UB_ALIGN_SIZE>（256B） |
workspace | 由 Host 侧SdmaWorkspaceManager分配的 GM 指针 |
syncId | MTE3/MTE2 管道同步事件 ID（0-7），避免与 kernel 内其他管道屏障冲突 |
baseConfig | {block_bytes, comm_block_offset, queue_num}，默认适用于单队列场景 |
channelGroupIdx | SDMA 通道组索引，默认使用get_block_idx()映射 |
URMA 构建（仅 Ascend950 / NPU_ARCH 3510）
bool BuildAsyncSession(__gm__ uint8_t *workspace, uint32_t destRankId, AsyncSession &session);异步约束
- 仅支持扁平连续的逻辑一维 tensor（非一维返回无效 event）
- SDMA workspace 必须由 Host 侧
SdmaWorkspaceManager分配 - URMA workspace 必须由 Host 侧
UrmaWorkspaceManager分配 - URMA 需要大页内存（
ACL_MEM_MALLOC_HUGE_ONLY），小页分配导致注册失败 scratchTile仅用于控制元数据，不是数据暂存缓冲
完成语义（Quiet 语义）
event.Wait(session)阻塞直到自上次 Wait 以来所有已发出的异步操作全部完成- 多次异步调用后，只需对最后一个
AsyncEvent调用一次Wait - 类似 shmem 的 quiet 语义
完整示例
// 构建会话 using ScratchTile = Tile<TileType::Vec, uint8_t, 1, comm::sdma::UB_ALIGN_SIZE>; ScratchTile scratchTile; TASSIGN(scratchTile, 0x0); comm::AsyncSession session; if (!comm::BuildAsyncSession<comm::DmaEngine::SDMA>(scratchTile, sdmaWorkspace, session)) { return; } // 批量传输 + 一次 Wait comm::AsyncEvent lastEvent; for (int rank = 0; rank < nranks; ++rank) { GT dstG(remoteDst + rank * size, shape, stride); lastEvent = comm::TPUT_ASYNC(dstG, srcG, session); } (void)lastEvent.Wait(session); // 等待所有 pending 操作完成【免费下载链接】pto-isaParallel Tile Operation (PTO) is a virtual instruction set architecture designed by Ascend CANN, focusing on tile-level operations. This repository offers high-performance, cross-platform tile operations across Ascend platforms.项目地址: https://gitcode.com/cann/pto-isa
创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
